Repository navigation
t27 worlds: refresh the catalog (14 repositories, 1737 specs) - #1176
Merged
Merged
Conversation
github-actions
Bot
force-pushed
the
bot/t27-world-scan
branch
from
September 26, 2026 08:50
97fd631 to
9af5b09
Compare
github-actions
Bot
force-pushed
the
bot/t27-world-scan
branch
2 times, most recently
from
September 28, 2026 10:05
fb22b38 to
aa38451
Compare
github-actions
Bot
force-pushed
the
bot/t27-world-scan
branch
from
September 29, 2026 10:06
aa38451 to
0dd0795
Compare
github-actions
Bot
force-pushed
the
bot/t27-world-scan
branch
from
September 30, 2026 09:57
0dd0795 to
39d396f
Compare
github-actions
Bot
force-pushed
the
bot/t27-world-scan
branch
from
October 1, 2026 10:25
39d396f to
ef78b36
Compare
gHashTag
added a commit
that referenced
this pull request
Oct 1, 2026
* feat(queen): one spec corpus for every tab of /game The owner's report: the specs were not in sync across the tabs of /game. Measured on t27.ai first: every tab happened to print 1577, because one scan wrote every file -- but the shell and the Spec Explorer frame each fetched the 1.8 MB manifest, the comb and the hero block each fetched the atlas, the SPECS rung read a sixth file, a tab switch dropped spec=, the comb's spec stayed local, and the Explorer ignored the world chosen in the shell. Nothing said two tabs were showing one corpus. - src/lib/queenCorpus.ts: one store for manifest, atlas, modules and foundation. Fetched once per window tree (the Explorer frame adopts the shell's registry), the wire first and the vendored file as fallback, each part saying its source and when it was generated; the manifest's identity is its commit and the SHA-256 of its bytes. Every tab reads it. - src/lib/queenCorpusCheck.ts: what "one corpus" means, as a pure function: manifest, atlas, shared core and the five spec-* ladders agree on count, on the manifest bytes they were built from, and world by world. - The address: a tab switch keeps spec= (and world=), the comb writes the spec it opens, and world= in the shell narrows the Explorer frame. - Shell root and Explorer root stamp data-spec-count / data-corpus-version. - qa/queen-spec-sync-contract.mjs (check:queen-spec-sync): files, source, store and address, each fed a broken fixture that must fail; --browser reads all 14 tabs of the built page. Wired into Website checks. - t27-world-scan.yml: the refresh PR it opens never got its required "T27 work report" (a GITHUB_TOKEN event starts no workflow), so #1176 (1737 specs) sat BLOCKED. It now dispatches pr-blog-report.yml for it; the discovery audit allows that dispatch and nothing else. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(qa): spec-sync browser mode starts Chrome on a Linux runner The CI runner's Chrome cannot start its sandbox; the other Queen browser contracts pass --no-sandbox --disable-dev-shm-usage on linux, and this one now does too (it exited 2, 'Chrome never announced a debugging port'). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
Oct 1, 2026
feat(queen): one spec corpus for every tab of /game (#1206) * feat(queen): one spec corpus for every tab of /game The owner's report: the specs were not in sync across the tabs of /game. Measured on t27.ai first: every tab happened to print 1577, because one scan wrote every file -- but the shell and the Spec Explorer frame each fetched the 1.8 MB manifest, the comb and the hero block each fetched the atlas, the SPECS rung read a sixth file, a tab switch dropped spec=, the comb's spec stayed local, and the Explorer ignored the world chosen in the shell. Nothing said two tabs were showing one corpus. - src/lib/queenCorpus.ts: one store for manifest, atlas, modules and foundation. Fetched once per window tree (the Explorer frame adopts the shell's registry), the wire first and the vendored file as fallback, each part saying its source and when it was generated; the manifest's identity is its commit and the SHA-256 of its bytes. Every tab reads it. - src/lib/queenCorpusCheck.ts: what "one corpus" means, as a pure function: manifest, atlas, shared core and the five spec-* ladders agree on count, on the manifest bytes they were built from, and world by world. - The address: a tab switch keeps spec= (and world=), the comb writes the spec it opens, and world= in the shell narrows the Explorer frame. - Shell root and Explorer root stamp data-spec-count / data-corpus-version. - qa/queen-spec-sync-contract.mjs (check:queen-spec-sync): files, source, store and address, each fed a broken fixture that must fail; --browser reads all 14 tabs of the built page. Wired into Website checks. - t27-world-scan.yml: the refresh PR it opens never got its required "T27 work report" (a GITHUB_TOKEN event starts no workflow), so #1176 (1737 specs) sat BLOCKED. It now dispatches pr-blog-report.yml for it; the discovery audit allows that dispatch and nothing else. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(qa): spec-sync browser mode starts Chrome on a Linux runner The CI runner's Chrome cannot start its sandbox; the other Queen browser contracts pass --no-sandbox --disable-dev-shm-usage on linux, and this one now does too (it exited 2, 'Chrome never announced a debugging port'). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
added a commit
that referenced
this pull request
Oct 1, 2026
- silicon coverage: the two vivado gf16 ports from #1176 consume GF16, they do not declare it; classified as EXCLUDED with the reason. - onboarding spec: counts follow the refreshed manifest (1737 specs, 14 repositories, 9 worlds); llms.txt / agents.t27 regenerated. - queen-contrast: IglaBoard.css registered in REACHED; the glance tiles low-pass the hex lattice with blur(6px) like every other board ground. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
added a commit
that referenced
this pull request
Oct 1, 2026
…1212) * feat(queen): IGLA board on WARS; black field; tables stack on phones The IGLA-Coder board (ternary 9M model writing t27 fn bodies, judged by t27c) becomes a section of the Queen WARS tab, in WARS style: - specs/queen/igla_board.t27 (module queen_igla_board) is the source; generated in gHashTag/igla-coder-gpu from committed artifacts. scripts/igla-board-from-spec.mjs recompiles it, re-derives every exact McNemar p from its discordants and writes the TS/JSON/public .t27. - IglaBoard: at-a-glance numbers, live jobs (read from the Railway board at view time, labelled SESSION-OBSERVED), mask A/B levers paired on two seeds, selection arena with gates and a negative control, pipeline, roadmap. - WARS field is black: green-tinted panels (rgba(1,16,13), #03120e, the inset green glow) are now black. - Every device: wide tables carry data-stack and per-cell data-label and become one card per row below 860px; the flow strip wraps. No sideways scroll at 320/768/1024/1440. - qa/igla-board-contract.mjs + CI step check:igla-board. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(queen): IGLA board adds the m14 lever (no effect, pooled p=0.5) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(website): green the checks after the 1737-spec catalog refresh - silicon coverage: the two vivado gf16 ports from #1176 consume GF16, they do not declare it; classified as EXCLUDED with the reason. - onboarding spec: counts follow the refreshed manifest (1737 specs, 14 repositories, 9 worlds); llms.txt / agents.t27 regenerated. - queen-contrast: IglaBoard.css registered in REACHED; the glance tiles low-pass the hex lattice with blur(6px) like every other board ground. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(wars): hide the stacked-table head with display:none on phones A 1px clipped thead kept its grid rows, and the viewport contract read that as an undeclared scroller at 390x844. Every cell already names its column through data-label. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
added a commit
that referenced
this pull request
Oct 1, 2026
…check:silicon check:silicon (Website checks) fails on main since #1176: gf16_dot4 and gf16_matmul_top name GF16 and are in no table. Both consume the format (a dot-product unit and the board top around it) and declare nothing, so they join the use/convert/measure exclusions with reasons. The world scan now runs check:silicon beside check:onboarding. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
Oct 1, 2026
feat(queen): IGLA board on WARS; black field; tables stack on phones (#1212) * feat(queen): IGLA board on WARS; black field; tables stack on phones The IGLA-Coder board (ternary 9M model writing t27 fn bodies, judged by t27c) becomes a section of the Queen WARS tab, in WARS style: - specs/queen/igla_board.t27 (module queen_igla_board) is the source; generated in gHashTag/igla-coder-gpu from committed artifacts. scripts/igla-board-from-spec.mjs recompiles it, re-derives every exact McNemar p from its discordants and writes the TS/JSON/public .t27. - IglaBoard: at-a-glance numbers, live jobs (read from the Railway board at view time, labelled SESSION-OBSERVED), mask A/B levers paired on two seeds, selection arena with gates and a negative control, pipeline, roadmap. - WARS field is black: green-tinted panels (rgba(1,16,13), #03120e, the inset green glow) are now black. - Every device: wide tables carry data-stack and per-cell data-label and become one card per row below 860px; the flow strip wraps. No sideways scroll at 320/768/1024/1440. - qa/igla-board-contract.mjs + CI step check:igla-board. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * feat(queen): IGLA board adds the m14 lever (no effect, pooled p=0.5) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(website): green the checks after the 1737-spec catalog refresh - silicon coverage: the two vivado gf16 ports from #1176 consume GF16, they do not declare it; classified as EXCLUDED with the reason. - onboarding spec: counts follow the refreshed manifest (1737 specs, 14 repositories, 9 worlds); llms.txt / agents.t27 regenerated. - queen-contrast: IglaBoard.css registered in REACHED; the glance tiles low-pass the hex lattice with blur(6px) like every other board ground. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(wars): hide the stacked-table head with display:none on phones A 1px clipped thead kept its grid rows, and the viewport contract read that as an undeclared scroller at 390x844. Every cell already names its column through data-label. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
gHashTag
added a commit
that referenced
this pull request
Oct 1, 2026
#1217) * fix(website): t27.ai stopped publishing when the catalog grew to 1737 specs trinity#1176 refreshed public/t27/manifest.json (1577 -> 1737 specs, 12 -> 14 repositories) and left specs/catalog/onboarding.t27 stating the old counts. onboarding-from-spec refuses a snapshot that disagrees with the manifest it ships beside, so `prebuild` on main exits 1 and the apex publisher (ghashtag.github.io, run 36893486012) failed at Build. - write the nine counts the generator names, and MEASURED_AT - correct the prose the new numbers made false: "every remaining fail never parsed" -- six now parse and lose only Verilog - regenerate /llms.txt and /agents.t27 from the spec - t27-world-scan runs check:onboarding with the other catalog gates, so a refresh that moves the counts stops before it commits instead of after it reaches main Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(website): classify the two gf16 ports #1176 vendored; scan gates check:silicon check:silicon (Website checks) fails on main since #1176: gf16_dot4 and gf16_matmul_top name GF16 and are in no table. Both consume the format (a dot-product unit and the board top around it) and declare nothing, so they join the use/convert/measure exclusions with reasons. The world scan now runs check:silicon beside check:onboarding. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
github-actions Bot
added a commit
that referenced
this pull request
Oct 1, 2026
fix(website): the catalog refresh runs the checks its counts must pass (#1217) * fix(website): t27.ai stopped publishing when the catalog grew to 1737 specs trinity#1176 refreshed public/t27/manifest.json (1577 -> 1737 specs, 12 -> 14 repositories) and left specs/catalog/onboarding.t27 stating the old counts. onboarding-from-spec refuses a snapshot that disagrees with the manifest it ships beside, so `prebuild` on main exits 1 and the apex publisher (ghashtag.github.io, run 36893486012) failed at Build. - write the nine counts the generator names, and MEASURED_AT - correct the prose the new numbers made false: "every remaining fail never parsed" -- six now parse and lose only Verilog - regenerate /llms.txt and /agents.t27 from the spec - t27-world-scan runs check:onboarding with the other catalog gates, so a refresh that moves the counts stops before it commits instead of after it reaches main Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * fix(website): classify the two gf16 ports #1176 vendored; scan gates check:silicon check:silicon (Website checks) fails on main since #1176: gf16_dot4 and gf16_matmul_top name GF16 and are in no table. Both consume the format (a dot-product unit and the board top around it) and declare nothing, so they join the use/convert/measure exclusions with reasons. The world scan now runs check:silicon beside check:onboarding. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
dmitrii-f-t27
added a commit
that referenced
this pull request
Oct 2, 2026
* fix(silicon): classify the four trinity-memory GF16 RTL specs The first catalog scan after dmitrii-f-t27/trinity-memory merged its GF16 RTL (issues #103-#113 there) stopped at check:silicon: gf16_codec, gf16_ffn, gf16_scalar and gf16_wide_norm name GF16 and sit in neither SPEC_TO_FORMAT nor EXCLUDED. Each says in its own header what it is -- a converter, an FFN controller, scalar arithmetic, a wide normalization unit -- and none declares the format, so all four go to EXCLUDED with that reason, beside the other GF16 consumers. The entries name paths the vendored corpus gains only in the same refresh, so this lands with the scan that vendors them, as #1176 did for its two gf16 ports. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * t27 worlds: refresh the catalog (14 repositories, 1771 specs) --------- Co-authored-by: Dmitrii Fedorov <dmitrii.f@t27.ai> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: t27 world scan <admin@t27.ai>
github-actions Bot
added a commit
that referenced
this pull request
Oct 2, 2026
t27 worlds: refresh the catalog (14 repositories, 1771 specs) (#1245) * fix(silicon): classify the four trinity-memory GF16 RTL specs The first catalog scan after dmitrii-f-t27/trinity-memory merged its GF16 RTL (issues #103-#113 there) stopped at check:silicon: gf16_codec, gf16_ffn, gf16_scalar and gf16_wide_norm name GF16 and sit in neither SPEC_TO_FORMAT nor EXCLUDED. Each says in its own header what it is -- a converter, an FFN controller, scalar arithmetic, a wide normalization unit -- and none declares the format, so all four go to EXCLUDED with that reason, beside the other GF16 consumers. The entries name paths the vendored corpus gains only in the same refresh, so this lands with the scan that vendors them, as #1176 did for its two gf16 ports. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> * t27 worlds: refresh the catalog (14 repositories, 1771 specs) --------- Co-authored-by: Dmitrii Fedorov <dmitrii.f@t27.ai> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: t27 world scan <admin@t27.ai>
10 of 11 tasks
dmitrii-f-t27
added a commit
that referenced
this pull request
Oct 10, 2026
The first catalog scan after dmitrii-f-t27/trinity-memory merged its GF16 RTL (issues #103-#113 there) stopped at check:silicon: gf16_codec, gf16_ffn, gf16_scalar and gf16_wide_norm name GF16 and sit in neither SPEC_TO_FORMAT nor EXCLUDED. Each says in its own header what it is -- a converter, an FFN controller, scalar arithmetic, a wide normalization unit -- and none declares the format, so all four go to EXCLUDED with that reason, beside the other GF16 consumers. The entries name paths the vendored corpus gains only in the same refresh, so this lands with the scan that vendors them, as #1176 did for its two gf16 ports.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A scheduled refresh of the .t27 catalog from GitHub. The scan, the vendor step, the shared core, the atlas and every catalog gate ran in
t27-world-scan.yml; this pull request carries only what they wrote underapps/website/public.1737 specs, 14 repositories — 160 added, 0 removed.
{ "version": 1, "head_sha": "ef78b36db2f3ab6eab9440679beb346dc25157df", "summary": "A scheduled scan of GitHub for .t27 specs refreshed the catalog behind the Queen map and the Spec Explorer: 1737 specs across 14 repositories, 160 added, 0 removed against the previous commit.", "changes": [ "The catalog now carries 1737 specs across 14 repositories (agent-mail-mcp, dmitrii-f-t27/trinity-memory, t27, tern-tc-9m-t27-verify, tern-tc-9m-verify, tri-net, trinity, trinity-fpga, trios, tt-trinity-corona, tt-trinity-euler, tt-trinity-gamma, tt-trinity-phi, turbobaby-user-bot); against the previous commit that is 160 added, 0 removed.", "Specs added, by directory: specs/port 122, dmitrii-f-t27/trinity-memory 15, trinity-fpga/specs 6, specs/automation 5, specs/api 4, specs/queen 2.", "Repositories new to the catalog in this run: tern-tc-9m-t27-verify, tern-tc-9m-verify.", "The shared core was rebuilt from this manifest and records 1737 specs and 126 module-name collisions.", "The universe atlas was rebuilt from that core and now describes 14 worlds and 1737 specs.", "Byte-identical copies discarded while vendoring, so one file does not appear twice in the library: 59." ], "tests": [ { "command": "npm run discover -- scan && npm run discover -- vendor", "result": "the GitHub scan qualified 14 repositories and vendored their .t27 files", "status": "passed", "evidence": "manifest.repos records t27@7696727e5, tri-net@e66972f8c, trinity-fpga@d7e9718e9, trinity@afc9d3843, tt-trinity-corona@2a5eefc43, dmitrii-f-t27/trinity-memory@096efbd01, agent-mail-mcp@a3b8f3d7f, tern-tc-9m-t27-verify@daf5f079e, tern-tc-9m-verify@885f9ccf8, trios@8b229e948, tt-trinity-euler@f07bc0815, tt-trinity-gamma@6fc94845d, tt-trinity-phi@1e51ee7a5, turbobaby-user-bot@55a666419" }, { "command": "npm run core -- index", "result": "the shared core was rebuilt from the manifest just written", "status": "passed", "evidence": "shared-core provenance.manifestSha256 bad7f5df3cc13af4 over 1737 specs" }, { "command": "npm run atlas -- scan && npm run atlas -- build", "result": "the map artifact was rebuilt from that same core", "status": "passed", "evidence": "universe-atlas at 2026-10-01T10:25:48.195Z with provenance.manifestSha256 bad7f5df3cc13af4" }, { "command": "npm run check:discovery check:shared-core check:spec-catalog check:queen-atlas check:queen-catalog check:queen-continuous check:queen-spec-mirrors check:skills-catalog check:crons-catalog check:agents check:tools check:docs", "result": "every catalog gate passed against the tree this pull request carries", "status": "passed", "evidence": "the workflow step \"Catalog gates\" ran them in order and the job reached the commit step, which only runs on success" }, { "command": "the vendored compiler wasm compiled every spec in the catalog", "result": "health across the corpus is 1308 ok, 254 warn, 175 fail", "status": "passed", "evidence": "manifest.health {\"ok\":1308,\"warn\":254,\"fail\":175} measured by 731581 bytes of t27_compiler.wasm" } ], "limitations": [ "Health is what the vendored compiler said at scan time, not a statement that every spec is correct or that its generated code was run.", "The scan reads public repositories only: a private repository writing .t27 specs is invisible to it and will not appear on the map.", "A spec that is byte-identical to one already vendored is discarded rather than listed twice, so a repository can qualify and still contribute nothing visible.", "9 t27 specs are not on the default ref and reach the site through a named overlay ref; a reader cloning master alone will not find them." ], "tags": [ "t27", "catalog", "automation", "specs" ], "blog": { "title": "A nightly scan keeps the spec catalog honest", "summary": "The map and the library are built from whatever GitHub holds tonight: 1737 specs across 14 repositories, vendored, compiled and gated before anything reaches the site.", "outline": [ "The catalog behind the map and the Spec Explorer is not a list somebody maintains by hand; it is the output of a scan that reads the public inventories of the owners a spec names and every repository a vendored source points at.", "This run ended with 1737 specs across 14 repositories (agent-mail-mcp, dmitrii-f-t27/trinity-memory, t27, tern-tc-9m-t27-verify, tern-tc-9m-verify, tri-net, trinity, trinity-fpga, trios, tt-trinity-corona, tt-trinity-euler, tt-trinity-gamma, tt-trinity-phi, turbobaby-user-bot), and against the previous commit that is 160 added, 0 removed — the numbers come from the artifacts themselves, not from the job's intentions.", "Each vendored file was compiled by the same wasm build of the compiler the browser runs, so the health shown beside a spec in the library is a measurement rather than a label: 1308 ok, 254 warn, 175 fail.", "The shared core and the universe atlas were rebuilt from this exact manifest in the same job, which is what keeps the map and the library describing one corpus instead of two snapshots taken at different times." ] } }